[HarfBuzz] finer-grained clustering options break behavior :(

[Jonathan Kew]

(oops, sent the original from the wrong email address so it probably 
didn't go through to the mailing list)

On 2/8/15 12:32, Jonathan Kew wrote:
> Hey Behdad,
>
> In https://bugzilla.mozilla.org/show_bug.cgi?id=729993, we'd like to 
> switch Gecko to use HB_BUFFER_CLUSTER_LEVEL_MONOTONE_CHARACTERS, so as 
> to enable independent mark coloring. However, it turns out this breaks 
> other aspects of rendering behavior; in particular, we get test 
> failures in a couple of tests using variation selectors.
>
> Looking at the code, I believe this is because _hb_ot_shape_normalize 
> relies on the .cluster field to identify the ranges within which it 
> operates. We'll need to fix this to use unicode properties instead, or 
> something like that.
>
> Similarly, we get a failure in a test for fallback mark positioning, 
> because _hb_ot_shape_fallback_position also relies on .cluster.
>
> For these operations, we need to work on a consistent "cluster" 
> regardless of the setting of the HB_BUFFER_CLUSTER_LEVEL option.
>
> JK
>
Just following up on this, see 
https://hg.mozilla.org/try/rev/b01499276dc2 for a patch I tried that 
fixes the test failures in Firefox.

JK

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/harfbuzz/attachments/20150802/d2f1df78/attachment.html>